Prediction of Ball Position in Three-Dimensional Space Using Artificial Neural Networks on Robot ERSOW

Abstract

The quality of the goalkeeping robots is essential to winning in the wheeled robots soccer competition. The goalkeeper robots should be able to block the arrival ball fast and correctly. This research aims to predict the ball position at the goal area in three-dimensional space using an Artificial Neural Network (ANN) and fusion camera on the ERSOW robot. The proposed system contributes to the development of the previous system by adding predictive capabilities from two-dimensional space into three dimensions. The system can estimate and predict the final position of the ball in three dimensional space. The dimensional space has been the x-axis, the y-axis, and the z-axis (height of the ball). Based on all the experiments, the estimation error value for the ball height or z-axis is 3.19 cm, the x-axis is 6.05 cm, and the y-axis is 9.98 cm. The prediction value for the rolling ball on the floor has the RMSE values in x-axis of 21.33 cm, y-axis of 1.29 cm, and z-axis of 0.04 cm. For the bouncing ball, it has a RMSE values in x-axis of 22.93 cm, y-axis of 1.37 cm, and z-axis of 1.40 cm.